163 lines
4.5 KiB
C++
163 lines
4.5 KiB
C++
#include "corner_harris.h"
|
|
|
|
#include "blur_gaussian.h"
|
|
#include "conversion_grayscale.h"
|
|
#include "edge_sobel.h"
|
|
#include <iostream>
|
|
|
|
CornerHarris::CornerHarris(PNM* img) :
|
|
Convolution(img)
|
|
{
|
|
}
|
|
|
|
CornerHarris::CornerHarris(PNM* img, ImageViewer* iv) :
|
|
Convolution(img, iv)
|
|
{
|
|
}
|
|
|
|
PNM* CornerHarris::transform()
|
|
{
|
|
int threshold = getParameter("threshold").toInt();
|
|
double sigma = getParameter("sigma").toDouble();
|
|
double sigma_weight = getParameter("sigma_weight").toDouble();
|
|
double k_param = getParameter("k").toDouble();
|
|
|
|
int width = image->width();
|
|
int height = image->height();
|
|
|
|
PNM* newImage = new PNM(width, height, QImage::Format_Mono);
|
|
|
|
math::matrix<float> Ixx(width, height);
|
|
math::matrix<float> Iyy(width, height);
|
|
math::matrix<float> Ixy(width, height);
|
|
|
|
this->corner_candidates = new math::matrix<float>(width, height);
|
|
this->corner_nonmax_suppress = new math::matrix<float>(width, height);
|
|
|
|
ConversionGrayscale* conversion_grayscale = new ConversionGrayscale(image);
|
|
PNM* gray_image = conversion_grayscale->transform();
|
|
|
|
BlurGaussian* blur_gaussian = new BlurGaussian(gray_image);
|
|
blur_gaussian->setParameter("size", 3);
|
|
blur_gaussian->setParameter("sigma", 1.6);
|
|
PNM* blur_gauss_image = blur_gaussian->transform();
|
|
|
|
EdgeSobel* edge_sobel = new EdgeSobel(blur_gauss_image);
|
|
math::matrix<float>* Gx = edge_sobel->rawHorizontalDetection();
|
|
math::matrix<float>* Gy = edge_sobel->rawVerticalDetection();
|
|
|
|
for (int i = 0; i < width; i++)
|
|
{
|
|
for (int j = 0; j < height; j++)
|
|
{
|
|
Ixx[i][j] = (*Gx)[i][j] * (*Gx)[i][j];
|
|
Ixy[i][j] = (*Gx)[i][j] * (*Gy)[i][j];
|
|
Iyy[i][j] = (*Gy)[i][j] * (*Gy)[i][j];
|
|
|
|
(*corner_candidates)[i][j] = 0;
|
|
(*corner_nonmax_suppress)[i][j] = 0;
|
|
}
|
|
}
|
|
|
|
for (int i = 1; i < width - 1; i++)
|
|
{
|
|
for (int j = 1; j < height - 1; j++)
|
|
{
|
|
float Sxx = 0;
|
|
float Syy = 0;
|
|
float Sxy = 0;
|
|
|
|
for (int k = -1; k <= 1; k++)
|
|
{
|
|
for (int l = -1; l <= 1; l++)
|
|
{
|
|
Sxx = Sxx + Ixx[i + k][j + l] * BlurGaussian::getGauss(k, l, sigma);
|
|
Syy = Syy + Iyy[i + k][j + l] * BlurGaussian::getGauss(k, l, sigma);
|
|
Sxy = Sxy + Ixy[i + k][j + l] * BlurGaussian::getGauss(k, l, sigma);
|
|
}
|
|
}
|
|
|
|
Sxx = Sxx / sigma_weight;
|
|
Sxy = Sxy / sigma_weight;
|
|
Syy = Syy / sigma_weight;
|
|
|
|
math::matrix<float> H(2,2);
|
|
|
|
H(0,0) = Sxx;
|
|
H(0,1) = Sxy;
|
|
H(1,0) = Sxy;
|
|
H(1,1) = Syy;
|
|
|
|
float detH = H(0, 0) * H(1, 1) - H(0, 1) * H(1, 0); //determinant
|
|
float trH = H(0, 0) + H(1, 1); //trace
|
|
float r = detH - k_param * pow(trH, 2);
|
|
|
|
if (r > threshold)
|
|
{
|
|
(*corner_candidates)[i][j] = r;
|
|
}
|
|
}
|
|
}
|
|
|
|
bool search = true;
|
|
|
|
while(search)
|
|
{
|
|
search = false;
|
|
for (int i = 1; i < width - 1; i++)
|
|
{
|
|
for (int j = 1; j <height - 1; j++)
|
|
{
|
|
float max = (*corner_candidates)[i][j];
|
|
|
|
for (int k = -1; k <= 1; k++)
|
|
{
|
|
for (int l = -1; l <= 1; l++)
|
|
{
|
|
if (max < (*corner_candidates)[i+k][j+l])
|
|
{
|
|
max = (*corner_candidates)[i+k][j+l];
|
|
}
|
|
}
|
|
}
|
|
|
|
if ((*corner_candidates)[i][j] == max)
|
|
{
|
|
(*corner_nonmax_suppress)[i][j] = (*corner_candidates)[i][j];
|
|
}
|
|
else
|
|
{
|
|
if ((*corner_candidates)[i][j] > 0)
|
|
{
|
|
search = true;
|
|
}
|
|
|
|
(*corner_nonmax_suppress)[i][j] = 0;
|
|
}
|
|
|
|
}
|
|
}
|
|
corner_candidates = corner_nonmax_suppress;
|
|
}
|
|
|
|
for (int i = 0; i < width; i++)
|
|
{
|
|
for (int j = 0; j < height; j++)
|
|
{
|
|
if ((*corner_candidates)[i][j] == 0)
|
|
{
|
|
|
|
newImage->setPixel(i, j, 0);
|
|
}
|
|
else
|
|
{
|
|
// std::cout << "White corner" << std::endl;
|
|
newImage->setPixel(i, j, 1);
|
|
}
|
|
}
|
|
}
|
|
return newImage;
|
|
}
|
|
|
|
|